loss = calc_loss_batch(input_batch, target_batch,model,
    device)

loss.backward()
optimizer.step()

        __
        __